Accurate histopathological image classification plays a crucial role in cancer detection and diagnosis. In automated cancer detection methods, extraction of histological features of malignant and benign tissues is a challenging task. This paper presents a modified approach on octave convolution to extract high and low-frequency features which help to provide a comprehensive representation of histopathological images. Proposed octave convolution model is used to perform histopathological image classification using three different optimization strategies. Firstly, an optimal alpha value of 0.5 is used to give equal importance to both high-frequency and low-frequency feature maps. This balanced approach ensures that the model effectively considers critical high-frequency features as well as low-frequency features of cancerous tissues. Secondly, high-frequency and low-frequency feature maps are extracted and down sampled into half the spatial dimension size to reduce the computational cost compared to standard CNN. Thirdly, training and validation was conducted using ReLU, PReLU, LeakyReLU, ELU, GELU and Swish activation functions. From the experiment, it was concluded that PReLU is the best activation function for capturing intricate patterns inherent in cancer-related histopathological images. Combining all these optimization strategies, the proposed method proved to provide a classification accuracy of 93% and also to reduce the computational cost by 50%. Performance validation against pre-trained models, CNN variants and vision transformer-based models has also been conducted, which proved superior performance of the proposed model.

Due to its influence in transcriptional potential of genes, genetic regulation by means of epigenetic mechanisms is essential for normal growth and development. In mammary cancer, epigenetic modifications play a key role for its development and progression. In early carcinogenesis stages, due to genetic alterations or environmental factors, chromatin structure alterations due to DNA methylation and post-translational modifications of DNA-bound proteins may appear. As with other types of tumor, genome-wide hypomethylation and hypermethylation of specific genes, particularly in CpG islands that normally are not methylated, are observed. In order to compare global DNA methylation levels between tumor tissue and normal mammary tissue, we studied 11 intact female dogs with mammary tumors. Both types of tissue were collected during surgery, with subsequent clinical staging and histopathological classification of tumors. For each animal, DNA was extracted from paired samples of tumor tissue and normal mammary tissue. Global genomic methylation levels were calculated by relative quantitation of 5-methyl-2’-deoxycytidine (5mdC) with HPLC. Results showed that tumoral tissue had a global DNA hypomethylation when compared with normal mammary tissue (P < 0.05). This difference was greater in high histopathological grade tumors, characterized by their aggressive clinical behavior and high metastatic rate. These findings underscore the importance of additional studies in this line of research, with greater sample sizes. In the future, global DNA methylation may be used as a prognosis biomarker for mammary cancer in dogs.

Background/Objectives: Breast cancer diagnosis based on histopathological examination remains the most reliable and widely accepted approach in clinical practice, despite being time-consuming and prone to inter-observer variability. While deep learning methods have achieved high accuracy in medical image classification, their cross-site generalization remains limited due to differences in staining protocols and image acquisition. This study aims to evaluate and compare three clinically relevant adaptation strategies to improve model robustness under domain shift. Methods: The ResNet50V2 model, pretrained on ImageNet and further fine-tuned on the Kaggle Breast Histopathology Images dataset, was subsequently adapted to the BreaKHis dataset under three clinically relevant transfer strategies: (i) threshold calibration without retraining (site calibration), (ii) head-only fine-tuning (light FT), and (iii) full fine-tuning (full FT). Experiments were performed on an internal balanced dataset and on the public BreaKHis dataset using strict patient-level splitting to avoid data leakage. Evaluation metrics included accuracy, precision, recall, F1-score, ROC-AUC, and PR-AUC, computed per magnification level (40×, 100×, 200×, 400×). Results: Full fine-tuning consistently yielded the highest performance across all magnifications, reaching up to 0.983 ROC-AUC and 0.980 sensitivity at 400×. At 40× and 100×, the model correctly identified over 90% of malignant cases, with ROC-AUC values of 0.9500 and 0.9332, respectively. Head-only fine-tuning led to moderate gains (e.g., sensitivity up to 0.859 at 200×), while threshold calibration showed limited improvements (ROC-AUC ranging between 0.60–0.73). Grad-CAM analysis revealed more stable and focused attention maps after full fine-tuning, though they did not always align with diagnostically relevant regions. Conclusions: Our findings confirm that full fine-tuning is essential for robust cross-site deployment of histopathology AI systems, particularly at high magnifications. Lighter strategies such as threshold calibration or head-only fine-tuning may serve as practical alternatives in resource-constrained environments where retraining is not feasible.

Malignant transformation of oral precancerous lesions is a multistep process intricately linked to the disruption of epithelial cell polarity and activation of the epithelial-mesenchymal transition (EMT) program. This study provides an integrated analysis of the polarity regulators PAR3, SCRIBBLE, and DLG7, elucidating their differential expression across normal oral mucosa (NOM), oral epithelial dysplasia (OED), and oral squamous cell carcinoma (OSCC). By combining histopathological evaluation, immunohistochemical profiling, and whole-transcriptome sequencing, this work offers novel insights into polarity disruption as a driving mechanism in oral tumorigenesis. Formalin-fixed, paraffin-embedded (FFPE) tissue specimens were obtained from 50 habitual tobacco users from West Bengal, India. Sections were stained with hematoxylin and eosin (H&E) for histopathological assessment and classification into normal oral mucosa (NOM), oral epithelial dysplasia (OED), and oral squamous cell carcinoma (OSCC) (well- and moderately-differentiated grades). Immunohistochemical (IHC) analysis was conducted to evaluate the expression and localization patterns of the polarity-associated proteins PAR3, SCRIBBLE, and DLG7. Complementing this, whole-transcriptome RNA sequencing was performed on biopsy specimens from an independent cohort of 25 oral cancer patients exhibiting both OED and OSCC lesions, enabling comparative gene expression profiling of the same polarity regulators. Statistical analysis using IBM SPSS (version 20.0) and Chi-square testing revealed a significant reduction or complete loss of PAR3, SCRIBBLE, and DLG7 expression in both oral epithelial dysplasia (OED) and oral squamous cell carcinoma (OSCC), compared to their moderate to strong expression in normal oral mucosa (NOM). This study reveals a striking decline in epithelial polarity protein expression from normal oral mucosa (NOM) to oral epithelial dysplasia (OED), followed by a modest resurgence in oral squamous cell carcinoma (OSCC). The strong concordance between immunohistochemical and transcriptomic profiles-with the exception of DLG7-highlights the disruption of cell polarity as an early and central molecular event in oral carcinogenesis. Collectively, the polarity regulators PAR3, SCRIBBLE, and DLG7 emerge as promising biomarkers for early malignant transformation in oral potentially malignant disorders (OPMDs) and as potential modulators of tumor initiation, progression, and invasive behavior.

Current histopathologic classifications do not reliably distinguish patients with primary cutaneous squamous cell carcinomas (cSCC) at risk of relapse from those with nonrelapsing tumors. This underscores the need for molecular signatures capable of stratifying patients during primary tumor resection. In this study, we used high-dimensional imaging mass cytometry and a 39-antibody panel to define the immune landscape of 20 primary cSCC with distinct clinical outcomes, four relapsing cSCC, and their perilesional skins. Computational analysis of spatially resolved single-cell data from 47 imaging mass cytometry images identified 12 immune-cell subsets that discriminated primary cSCC from perilesional skin. Regulatory T cells, cytotoxic CD8+ T lymphocytes, and tumor-associated macrophages and neutrophils characterized tumors, whereas Langerhans cells and skin-resident macrophages defined perilesional skin. Skin-resident macrophages were characterized by the expression of CD206, CD11c, and HLA-DR and the absence of CD14. These cells infiltrated tumors from nonrelapsing patients more efficiently. We found a higher density of proliferating, mature, and cytotoxic cells within this macrophage subset, consistent with the absence of relapse. Spectral flow cytometry analysis on fresh tumor biopsies revealed that the skin-resident macrophages had phagocytic properties, suggesting a role in tumor antigen processing. Additionally, neighborhood profiling revealed that DC-LAMP+ dendritic cells were in close proximity with helper and cytotoxic T lymphocytes in primary cSCC from patients without relapse, indicative of active adaptive immunity. Our findings identify phagocytic skin-resident macrophages and dendritic cell-T cell clusters as features differentiating nonrelapsing cSCC from primary cSCC at risk of relapse. These data have the potential to guide the identification of prognostic biomarkers for cSCC.

For patients with acute lymphoblastic leukemia (ALL), one of the main causes of cancer-related mortality, a timely and precise diagnosis is essential for improving their prognosis. To achieve this, this paper presents a sequential deep learning method for the classification of ALL based on the histopathological diagnosis of PBS images. The publicly accessible Kaggle dataset was used to extract image samples from 3256 benign patients and three types of malignancy (Initial Pre-B, Intermediate Pre-B, and Advanced Pro-B). Using data augmentation techniques, the database's size was increased to 6,512 photos to make the model more broadly applicable. After individual training and evaluation, the five pre-trained deep learning models—InceptionNetV3, EfficientNetB0, VGG19, ResNet50, and DenseNet201—achieved accuracy rates of 93.2%, 92.5%, 91.8%, 90.3%, and 89.7%, respectively. The models' overall accuracy for a hierarchical class was evaluated at an astounding 98.15%. The performance evaluation indicates that the model is adjustable with an MCC of 0.973 and a Kappa of 0.97. In clinical use, the new approach significantly decreased the misclassification rate and outperformed the single models, indicating that it may be a dependable and effective diagnostic method for early detection of leukemia.

Imaging of Gynecologic Neuroendocrine Tumors: A Case-Based Pictorial Essay.

Integrating artificial intelligence-driven digital pathology and genomics to establish patient-derived organoids as new approach methodologies for drug response in head and neck cancer.

Automated Kidney Tumor Segmentation in CT Images Using Deep Learning: A Multi-Stage Approach.

arly and accurate detection of cancer, via histopathology, can improve patient outcomes significantly. However, manual analysis of microscopic tissue images is time-consuming and subjective. Machine learning (ML), especially deep learning (DL), provides automated solutions to classify cancer cell images. In this review, we summarize state-of-the-art ML and DL techniques for histopathological image classification. We discuss common public datasets (e.g. BreakHis), preprocessing steps (e.g. stain normalization, data augmentation), and modern model architectures (e.g. CNNs like ResNet, EfficientNet, Vision Transformers). Evaluation typically uses metrics like accuracy, precision, recall, F1-score, and ROC-AUC. For example, a transfer-learning approach with ResNet-50 achieved 92.42% accuracy (AUC 0.9986) on an 8-class breast cancer subtyping task [1]. Other hybrid models have reported accuracies up to 99% on binary classification of breast histology [2]. However, there are challenges with overfitting and interpretability. We will conclude by outlining future directions for cancer image classification.

Neural networks have achieved state-of-the-art performance in classifying whole slide image (WSI) patches in histopathology through supervised learning. However, their reliance on large-scale annotated datasets imposes a substantial labeling burden, limiting the practical benefits of AI-assisted diagnostics. Foundation models, pretrained on diverse datasets for multi-purpose applications, offer a promising alternative by enabling out-of-the-box generalization. Despite their success in other domains, these models currently underperform in histopathology due to domain-specific challenges. In this work, we introduce a fine-tuning pipeline that significantly enhances the performance of foundation models for histopathological classification using only a minimal amount of labeled data. Specifically, we curate an unlabeled dataset from the target domain and employ self-supervised learning (SSL) to adapt pretrained Vision Transformers (ViTs). Our approach substantially improves classification accuracy while reducing annotation requirements, making foundation models more suitable for histopathological analysis. Furthermore, our results show that SSL-trained models can extract richer features even without access to class labels or balanced training data.

Background/Objectives: Accurate histopathological classification of lung and colon tissues remains difficult due to subtle morphological overlap between benign and malignant regions. Deep learning approaches have advanced diagnostic precision, yet models often lack interpretability or require complex multi-stage pipelines. This study aimed to develop an end-to-end dual-branch attention network capable of achieving high accuracy while preserving computational efficiency and transparency. Methods: The architecture integrates EfficientNetV2-B0 and MobileNetV3-Small backbones through a cross-gated fusion mechanism that adaptively balances global context and fine structural details. Efficient channel attention and generalized mean pooling enhance discriminative learning without external feature extraction or optimization stages. Results: The network achieved 99.84% accuracy, precision, recall, and F1-score, with an MCC of 0.998. Grad-CAM maps showed strong spatial correspondence with diagnostically relevant histological structures. Conclusions: The end-to-end framework enables the reliable, interpretable, and computationally efficient classification of lung and colon histopathology and has potential applicability to computer-assisted diagnostic workflows.

INTRODUCTION Over the past decade, the pathology of un differentiated sinonasal malignancies has undergone extensive study, leading to significant advancements in the depiction and histopathological classification of various entities. These entities are now recognized as subsets of "sinonasal undifferentiated carcinomas (SNUC)" and poorly differentiated unclassified carcinomas. Typically, these malignancies are detected at later stages, by which time they have often invaded the facial and cranial regions. Olfactory neuroblastoma, which arises from the olfactory neuroepithelium with neuroblastic immature differentiation, is one such malignancy. CASE REPORT We present a case involving a left-sided nasal mass with blood-tinged discharge that obscured the nasal cavity. Previous biopsies had been inconclusive. Imaging revealed a large heterogeneous mass with bone erosions and extension into the intraorbital and intracranial regions. The patient underwent an endoscopic nasal biopsy. CONCLUSION Olfactory Neuroblastoma is a rare ,highly malignant, often have a long history before diagnosis. Treatment utilises a combination of surgery, external beam radiation, and chemotherapy modalities. Immunohistochemistry plays a crucial role in establishing a definitive diagnosis.

Correct histopathological image classification of lung and colon cancer is a stringent challenge for clinical pathology. This work introduces a hybrid deep learning network by combining traditional handcrafted features of LBP, GLCM, wavelet, color, and morphological descriptors with deep features derived from an extended EfficientNetB0. A transformer-based attention fusion strategy is adopted to fuse these heterogeneous representations, facilitating robust multi-scale feature learning. To even better accommodate adaptability and curtail catastrophic forgetting, the model is trained with an adaptive incremental learning approach with stage-wise data augmentation. The suggested method is trained on the LC25000 dataset and tested on two public, independent datasets, NCT-CRC-HE-100K and HMU-GC-HE-30K, showing consistent performance with accuracies of 99.87%, 99.07%, and 98.4%, respectively. These findings are affirmations of the framework's generalizability, scalability, and clinical applicability in multi-class histopathological image classification. All source code and dataset access instructions are publicly made available to encourage reproducibility and extension.

Cancer remains one of the leading causes of global mortality, with lung, colon, skin, and breast cancers contributing significantly to the disease burden. Accurate and timely classification of histopathological images is critical for effective diagnosis and treatment planning. However, existing deep learning models for histopathology often achieve strong results but remain limited to single-cancer classification, lack generalizability across datasets, and provide little transparency for clinical use. To address these gaps, we propose CancerDet-Net, a comprehensive unified framework capable of classifying nine histopathological subtypes across four major cancer types. CancerDet-Net integrates separable convolutional layers, Vision Transformer (ViT) blocks with local-window sparse self-attention, and a Hierarchical Multi-Scale Gated Attention Mechanism (HMSGA), combined through Cross-Scale Feature (CSF) Fusion. Unlike prior approaches, our model not only achieves top-performing accuracy (98.51%) but also incorporates explainable AI (XAI) visualizations and is deployed via both a web-based platform and Android app for real-time clinical use. This combination of multi-cancer generalization, interpretability, and deployment readiness establishes CancerDet-Net as a distinctive contribution to AI-driven digital pathology.

Lung cancer remains a leading cause of cancer-related mortality, primarily due to diagnostic inconsistencies and limitations of conventional methods. This study addresses the critical need for accurate, transparent, and clinically viable diagnostic systems by proposing a novel deep learning framework for histopathological lung cancer classification. Our research introduces a hybrid ensemble architecture that combines the hierarchical feature extraction capabilities of ResNet-50 with the global contextual understanding of Vision Transformer (ViT). Input images are processed in parallel through both pathways: ResNet-50 extracts 2048-dimensional spatial features via convolutional and residual blocks followed by global average pooling, while ViT generates 768-dimensional features from patch embeddings and a transformer encoder. These features are then fused into a 2816-dimensional combined vector, which is fed into a classification head comprising three fully connected layers with Batch Normalization, ReLU activation, and Dropout regularization, culminating in a 3-class softmax output. The ensemble model demonstrated superior performance, achieving a mean cross-validation accuracy of 99.96% ± 0.0004%, a holdout test set accuracy of 99.94%, and a separate test set accuracy of 99.82%. Furthermore, the integration of a multi-disciplinary Explainable AI (XAI) strategy, including Grad-CAM, LIME, SHAP, Saliency Maps, Integrated Gradients, and Occlusion Sensitivity, provided crucial interpretability, with attention heatmaps showing 87.3% overlap with pathologist-identified regions of interest. This work significantly advances AI-assisted lung cancer diagnosis by offering a robust, highly accurate, and interpretable solution that addresses the current clinical gaps and holds huge potential for improving patient outcomes.

This study aimed to develop a deep learning model to recognize cell interaction patterns in pathological slides of malignant bowel obstruction. The model classifies lesions into four categories—normal mucosa, serrated lesions, adenomas, and adenocarcinomas—and evaluates its diagnostic utility in tumor-associated obstruction. Pathological slides from patients with tumor-induced intestinal obstruction (TICO) were retrospectively collected from First Affiliated Hospital of Bengbu Medical University and annotated into four histological categories: normal, serrated lesions, adenomas, and adenocarcinomas. The proposed deep learning framework combines a residual convolutional network with a bidirectional state-space module (SSM), enabling multiscale feature extraction through convolution and down-sampling, while modeling the spatiotemporal dynamics of cellular interactions. The model was designed to learn spatial and structural characteristics of cell interactions—such as glandular organization, intercellular spacing, and nuclear density—across different lesion types. Grad-CAM was used to visualize attention regions and assess consistency between model focus and pathological features. However, Grad-CAM was used solely for interpretability and not clinical validation; no expert verification of the visualizations has been performed. On an independent Chaoyang test set, the model achieved a validation accuracy of 85% and a macro-F1 score of 0.843 (95% CI: 0.829–0.857), showing only a 3% decline from training accuracy (88%), thus demonstrating strong generalizability. In addition, we calculated 95% confidence intervals using 1,000 bootstrap resamples and applied both the DeLong test and McNemar test to compare the performance of our model with baseline methods. The results demonstrated statistically significant improvements (P < 0.05) in Accuracy, Macro-F1, and ROC-AUC, thereby further strengthening the reliability of our conclusions. The recall for adenocarcinoma (Class 3) reached 88%, while Classes 0–2 (normal, serrated lesions, and adenomas) ranged from 78% to 83%. These results highlight the impact of sample imbalance and morphological similarity, which will be addressed in future work through Focal Loss reweighting and detailed error analysis. Grad-CAM visualizations identified regions of glandular disruption and abnormal nuclear density, aligning with WHO-2022 diagnostic criteria and enhancing model interpretability. Overall performance is comparable to state-of-the-art gastrointestinal pathology AI systems from recent years, offering rapid and quantitative diagnostic support in emergency pathology settings. The proposed deep learning model effectively distinguishes four categories of tumor-associated colorectal lesions, demonstrating strong diagnostic potential. Limitations include: (i) all data were retrospectively collected from a single center, without external multicenter validation. Differences in population composition, scanning platforms, and staining batches may affect the model’s external generalizability; future studies will prioritize the inclusion of multicenter datasets to systematically evaluate the robustness and applicability of the model under diverse clinical conditions; (ii) the model has so far been assessed only in an offline environment, lacking prospective clinical validation within real-world workflows. Nonetheless, this model provides an important foundation for the early diagnosis of TICO, the formulation of personalized treatment strategies, and the advancement of pathological image analysis technologies.

Abstract BACKGROUND Dysembryoplastic neuroepithelial tumor (DNT) is a rare glioneuronal tumor that typically arises in childhood and is strongly associated with drug-resistant epilepsy. Diagnosis is often challenging due to histological heterogeneity. Therefore, DNT may be an ideal candidate for genome-wide DNA methylation profiling. However, integrated analyses combining histopathology and genome-wide DNA methylation profiling remain limited. METHODS We retrospectively analyzed the clinical data, histopathological diagnoses, and molecular findings of patients diagnosed with DNT or low-grade glioneuronal tumors mimicking DNT at the Japan Children’s Cancer Group Central Diagnosis between February 2016 and March 2025. RESULTS Twenty-three patients (12 males and 11 females) were included. The median age at the diagnosis was 11 years (range, 3 to 24 years). The majority were located in the temporal lobes (n=12) and other cortical regions (n=10), but a case was located in the third ventricle. Seizure onset was noted in 15 cases. FGFR1 tyrosine kinase domain internal tandem duplication was identified in 10 cases, and FGFR1 K656E mutation in 5 cases. Genome-wide DNA methylation profiling was available in 16 cases. According to the DKFZ classifier, 14 cases were classified as DNT, one as Pilocytic astrocytoma (PA), hemispheric, and one as Glioneuronal tumour, subtype A (novel). A case of PA, hemispheric was histopathologically diagnosed as DNT complex form and showed focal areas with PA morphology. Interestingly, a case of Glioneuronal tumour, subtype A (novel) showed atypical clinical features such as non-seizure onset and location in the third ventricle. It showed recurrence 5 years after an initial surgery with histopathological changes, including loss of specific glioneuronal elements and growth of multinucleated cells. CONCLUSIONS Discrepancy between histopathological diagnosis and DNA methylation classification may result from varied cellular composition. In addition, atypical DNA methylation classification may be associated with clinical signatures, requiring integrated analyses in larger cohorts.

Postmortem microbiological tests can help confirm infectious causes of death and identify the responsible microorganism (MO). However, challenges arise due to the lack of consensus and regulatory guidelines. This prospective single-center study conducted at University Hospitals Leuven from 2013 to 2016, aimed to assess the relevance as well as site- and time-related aspects of postmortem microbiology. A total of 200 clinical and forensic autopsies were included in the study, with 1321 samples collected. Approximately one-fifth of cases revealed an infectious cause of death, predominantly pneumonia. Histopathological classification into infectious and non-infectious causes of death served as the gold standard against which microbiological results were compared. The group of MO (i.e., group 1 and group 2 pathogens as well as a small number of specific pathogens), the number of positive sites, specific sampling sites, and the unique presence of a pathogen at a given site were all significantly related to infectious deaths. We propose that the concept of polymicrobial overgrowth as contamination primarily pertains to non-pathogenic MOs. We recommend the sampling of internal or external peripheral blood, heart blood, left lung, and spleen as part of a standardized protocol. Although the postmortem interval (PMI) was shown to affect microbiological results, its clinical relevance warrants further investigation.

Evaluating the impact of omitting contralateral systematic prostate biopsies on histopathological grading, risk classification, and lymph node involvement probability in patients with unilateral suspicious prostate cancer on magnetic resonance imaging